Multimodel_setup / scripts /package_and_upload_hf_dataset.py
Thiru1122's picture
feat(backup): add Hugging Face dataset packaging and upload pipeline script
ae17379
Raw History Blame Contribute Delete
8.82 kB
"""
Package and Upload Full Dataset and Legacy Project to Private Hugging Face Dataset Repo
Repo: https://huggingface.co/datasets/Thiru1122/deepfake-dataset-full
"""
import os
import sys
import time
import zipfile
import hashlib
import shutil
from pathlib import Path
from huggingface_hub import HfApi
REPO_ID = "Thiru1122/deepfake-dataset-full"
REPO_TYPE = "dataset"
SOURCE_ROOT = Path(r"C:\Users\user\Desktop\multi model connection\currently working")
STAGING_DIR = Path(r"C:\Users\user\Desktop\hf_dataset_staging")
def compute_sha256(filepath: Path) -> str:
hasher = hashlib.sha256()
with open(filepath, "rb") as f:
while chunk := f.read(1024 * 1024):
hasher.update(chunk)
return hasher.hexdigest()
def create_zip_archive(source_dir: Path, output_zip: Path, arc_prefix: str = ""):
print(f"[{time.strftime('%X')}] Archiving {source_dir} -> {output_zip.name}...")
start_time = time.time()
file_count = 0
with zipfile.ZipFile(output_zip, "w", compression=zipfile.ZIP_DEFLATED, compresslevel=1) as zf:
for root, _, files in os.walk(source_dir):
for file in files:
full_path = Path(root) / file
rel_path = full_path.relative_to(source_dir)
archive_name = f"{arc_prefix}/{rel_path}" if arc_prefix else str(rel_path)
zf.write(full_path, archive_name)
file_count += 1
if file_count % 20000 == 0:
print(f" ... archived {file_count:,} files ({time.time() - start_time:.1f}s)")
duration = time.time() - start_time
size_mb = output_zip.stat().st_size / (1024 * 1024)
print(f"[{time.strftime('%X')}] Done: {output_zip.name} ({file_count:,} files, {size_mb:.2f} MB in {duration:.1f}s)")
return file_count, size_mb
def create_project_zip(source_dir: Path, output_zip: Path):
print(f"[{time.strftime('%X')}] Archiving project source files -> {output_zip.name}...")
start_time = time.time()
file_count = 0
# Exclude Dataset directory since it is archived separately
with zipfile.ZipFile(output_zip, "w", compression=zipfile.ZIP_DEFLATED, compresslevel=3) as zf:
for root, dirs, files in os.walk(source_dir):
rel_root = Path(root).relative_to(source_dir)
# Skip Dataset and .git directories
if rel_root.parts and rel_root.parts[0] in ("Dataset", ".git"):
continue
for file in files:
full_path = Path(root) / file
rel_path = full_path.relative_to(source_dir)
zf.write(full_path, str(rel_path))
file_count += 1
duration = time.time() - start_time
size_mb = output_zip.stat().st_size / (1024 * 1024)
print(f"[{time.strftime('%X')}] Done: {output_zip.name} ({file_count:,} files, {size_mb:.2f} MB in {duration:.1f}s)")
return file_count, size_mb
def main():
print(f"=== Starting Hugging Face Dataset Backup to {REPO_ID} ===")
STAGING_DIR.mkdir(parents=True, exist_ok=True)
api = HfApi()
# 1. Ensure private repository exists
repo_url = api.create_repo(
repo_id=REPO_ID,
repo_type=REPO_TYPE,
private=True,
exist_ok=True,
)
print(f"Repository verified: {repo_url}")
checksums = {}
# 2. Archive Dataset Splits
dataset_dir = SOURCE_ROOT / "Dataset"
splits = [
("Train", "Dataset_Train.zip", "Train"),
("Validation", "Dataset_Validation.zip", "Validation"),
("Test", "Dataset_Test.zip", "Test"),
]
for folder_name, zip_name, arc_prefix in splits:
split_path = dataset_dir / folder_name
zip_path = STAGING_DIR / zip_name
if not zip_path.exists():
create_zip_archive(split_path, zip_path, arc_prefix)
else:
print(f"Archive {zip_name} already exists. Skipping packaging.")
sha = compute_sha256(zip_path)
checksums[zip_name] = (sha, zip_path.stat().st_size)
print(f" SHA-256: {sha}")
# 3. Archive Full Project (Excluding Dataset)
project_zip = STAGING_DIR / "currently_working_project.zip"
if not project_zip.exists():
create_project_zip(SOURCE_ROOT, project_zip)
else:
print(f"Archive {project_zip.name} already exists. Skipping packaging.")
sha_proj = compute_sha256(project_zip)
checksums[project_zip.name] = (sha_proj, project_zip.stat().st_size)
print(f" SHA-256: {sha_proj}")
# 4. Copy standalone manifest CSVs directly
manifest_csv = SOURCE_ROOT / "dataset_manifest.csv"
duplicates_csv = SOURCE_ROOT / "duplicates_metadata.csv"
staged_manifest = STAGING_DIR / "dataset_manifest.csv"
staged_duplicates = STAGING_DIR / "duplicates_metadata.csv"
if manifest_csv.exists() and not staged_manifest.exists():
shutil.copy2(manifest_csv, staged_manifest)
if duplicates_csv.exists() and not staged_duplicates.exists():
shutil.copy2(duplicates_csv, staged_duplicates)
for staged_file in [staged_manifest, staged_duplicates]:
if staged_file.exists():
sha = compute_sha256(staged_file)
checksums[staged_file.name] = (sha, staged_file.stat().st_size)
print(f"Copied {staged_file.name} | SHA-256: {sha}")
# 5. Generate Checksums File
checksum_file = STAGING_DIR / "checksums.sha256"
with open(checksum_file, "w") as f:
for fname, (sha, size) in checksums.items():
f.write(f"{sha} {fname} ({size:,} bytes)\n")
# 6. Generate Dataset README.md
readme_path = STAGING_DIR / "README.md"
readme_content = f"""---
license: mit
task_categories:
- image-classification
size_categories:
- 100K<n<1M
tags:
- deepfake
- face-manipulation
- dataset-backup
---
# Deepfake Detection Clean Dataset & Project Backup
Private lossless archive of the deepfake image dataset and associated project assets.
## Dataset Summary
- **Total Images:** 190,335
- **Train Split:** 140,002 images (`Dataset_Train.zip`)
- **Validation Split:** 39,428 images (`Dataset_Validation.zip`)
- **Test Split:** 10,905 images (`Dataset_Test.zip`)
- **Manifest:** `dataset_manifest.csv` (19.8 MB)
- **Duplicates Metadata:** `duplicates_metadata.csv` (5.2 MB)
- **Legacy Project Code & Models:** `currently_working_project.zip` (Flask app, LSTM/ResNet `df_model.pt`, notebooks, test videos)
## File Integrity & Checksums (SHA-256)
```text
"""
for fname, (sha, size) in checksums.items():
readme_content += f"{sha} {fname} ({size / (1024*1024):.2f} MB)\n"
readme_content += """```
## How to Restore
```python
from huggingface_hub import hf_hub_download
import zipfile
for zip_name in ["Dataset_Train.zip", "Dataset_Validation.zip", "Dataset_Test.zip"]:
local_path = hf_hub_download(
repo_id="Thiru1122/deepfake-dataset-full",
repo_type="dataset",
filename=zip_name,
)
with zipfile.ZipFile(local_path, "r") as zf:
zf.extractall("Dataset")
```
"""
with open(readme_path, "w", encoding="utf-8") as f:
f.write(readme_content)
print(f"Generated {readme_path.name}")
# 7. Upload All Files to Hugging Face
files_to_upload = [
"README.md",
"checksums.sha256",
"dataset_manifest.csv",
"duplicates_metadata.csv",
"Dataset_Test.zip",
"Dataset_Validation.zip",
"Dataset_Train.zip",
"currently_working_project.zip",
]
print("\n=== Commencing Uploads to Hugging Face ===")
for fname in files_to_upload:
fpath = STAGING_DIR / fname
if not fpath.exists():
print(f"Warning: {fpath} does not exist. Skipping.")
continue
size_mb = fpath.stat().st_size / (1024 * 1024)
print(f"[{time.strftime('%X')}] Uploading {fname} ({size_mb:.2f} MB)...")
upload_start = time.time()
api.upload_file(
path_or_fileobj=str(fpath),
path_in_repo=fname,
repo_id=REPO_ID,
repo_type=REPO_TYPE,
commit_message=f"Upload {fname} (SHA-256: {checksums.get(fname, ('', ''))[0][:8]})",
)
upload_duration = time.time() - upload_start
print(f"[{time.strftime('%X')}] Uploaded {fname} in {upload_duration:.1f}s")
# 8. Verify Files on Hugging Face
print("\n=== Verifying Files on Remote Repository ===")
remote_files = api.list_repo_files(repo_id=REPO_ID, repo_type=REPO_TYPE)
for rf in remote_files:
print(f" [Remote Verified] {rf}")
print(f"\nAll files successfully uploaded to: https://huggingface.co/datasets/{REPO_ID}")
if __name__ == "__main__":
main()